将 stderr 记录到文件，以日期时间为前缀

Question

. Answers to this question are eligible for a +100 reputation bounty. Basj is looking for a .

I do proper logging with the logging module ( logger.info , logger.debug ...) and this gets written to a file.

But in some corner cases (external modules, uncaught exceptions, etc.), I sometimes still have errors written to stderr .

I log this to a file with:

import sys
sys.stdout, sys.stderr = open("stdout.log", "a+", buffering=1), open("stderr.log", "a+", buffering=1)
print("hello")
1/0

It works, but how to also have the datetime logged before each error?

Note: I'd like to avoid to use logging for this part, but something more low level.

I also want to avoid this solution:

def exc_handler(ex_cls, ex, tb):
    with open('mylog.log', 'a') as f:
        dt = time.strftime('%Y-%m-%d %H:%M:%S')
        f.write(f"{dt}\n")
        traceback.print_tb(tb, file=f)
        f.write(f"{dt}\n")

sys.excepthook = exc_handler

because some external modules might override this. Is there a low level solution like overriding sys.stderr.print ?

Answer 1

You can use the approach described in this cookbook recipe to treat a logger like an output stream, and then redirect sys.std{out,err} to log stuff written to them. Then, of course, you can configure logging to use whatever format you want, including a date-time prefix, in the normal way using eg %(asctime)s in the format.

Answer 2

For the benefit of random readers, if you are simply looking to "log stderr to file" - this is not the answer you are looking for. Look at python's logging module; that is the best approach for most use-cases.
With the "disclaimer" part over-with, a lower-level approach to sys.std%s is very possible, but might be quite tricky. The following example:

1. Is based on the sys.stdout, sys.stderr = open(...) approach from the question itself.
2. Properly handles most cases of partial input (at least those I could think of) - not all std writes end with a newline.
3. Properly handles newlines (as far as I could test).
4. Does not propagate to its children.
5. Is extremely difficult to debug when things go wrong (and oh-boy; so many things went wrong for me while writing this..).

#!/usr/bin/env python
import sys
import os
import datetime
import time
import subprocess
from types import MethodType

def get_timestamp_prefix():
    return f"{datetime.datetime.now()!s}: "


def stdwriter(self, msg, *args, **kwargs):
    tstamp_prefix = get_timestamp_prefix()

    if self._last_write_ended_with_newline:
        # Add current timestamp if we're starting a new line
        self._orig_write(tstamp_prefix)
    self._last_write_ended_with_newline = msg.endswith('\n')

    if msg.endswith('\n'):
        # Avoid associating current timestamp to next error message
        self._orig_write(f"\n{tstamp_prefix}".join(msg.split('\n')[:-1]))
        self._orig_write("\n")
    else:
        self._orig_write(f"\n{tstamp_prefix}".join(msg.split('\n')))


def setup_std_files():
    sys.stdout, sys.stderr = open("stdout.log", "a+", buffering=1), open("stderr.log", "a+", buffering=1)
    for stream_name in ("stdout", "stderr"):
        sys_stream = getattr(sys, stream_name)
        setattr(sys_stream, "_last_write_ended_with_newline", True)
        setattr(sys_stream, "_orig_write", getattr(sys_stream, "write"))
        setattr(sys_stream, "write", MethodType(stdwriter, sys_stream))


def print_some_stuff():
    print("hello")
    print("world")
    print("and..,", end=" ")
    time.sleep(2.5)
    # demonstrating "single" (close enough) timestamp until a newline is encountered
    print("whazzzuppp?")
    print("this line's timestamp should be ~2.5 seconds ahead of 'and.., whazzzuppp'")


def run_some_failing_stuff():
    try:
        1/0
    except (ZeroDivisionError, subprocess.CalledProcessError) as e:
        print(f"Exception handling: {e!r}") # to STDOUT
        raise e                             # to STDERR (only after `finally:`)
    else:
        print("This should never happen")
    finally:
        print("Getting outta' here..")  # to STDOUT


if __name__ == "__main__":
    setup_std_files()
    print_some_stuff()
    run_some_failing_stuff()

Running this code will output:

$ rm *.log; ./pyerr_division.py ; for f in *.log; do echo "====== $f ====="; cat $f; echo "====== end ====="; done
====== stderr.log =====
2023-01-14 06:59:02.852233: Traceback (most recent call last):
2023-01-14 06:59:02.852386:   File "/private/tmp/std_files/./pyerr_division.py", line 63, in <module>
2023-01-14 06:59:02.853152:     run_some_failing_stuff()
2023-01-14 06:59:02.853192:   File "/private/tmp/std_files/./pyerr_division.py", line 53, in run_some_failing_stuff
2023-01-14 06:59:02.853294:     raise e                             # to STDERR (only after `finally:`)
2023-01-14 06:59:02.853330:   File "/private/tmp/std_files/./pyerr_division.py", line 50, in run_some_failing_stuff
2023-01-14 06:59:02.853451:     1/0
2023-01-14 06:59:02.853501: ZeroDivisionError: division by zero
====== end =====
====== stdout.log =====
2023-01-14 06:59:00.346447: hello
2023-01-14 06:59:00.346502: world
2023-01-14 06:59:00.346518: and.., whazzzuppp?
2023-01-14 06:59:02.851982: this line's timestamp should be ~2.5 seconds ahead of 'and.., whazzzuppp'
2023-01-14 06:59:02.852039: Exception handling: ZeroDivisionError('division by zero')
2023-01-14 06:59:02.852077: Getting outta' here..
====== end =====

---

Child propagation is not explicitly in-scope for this question. Still, changing the approach not only allows to collect children's std%s to the same files, but also provides for easier debugging. The idea is to write to the original std%s if problems occur.
The following is based on this answer (thanks user48..2):

def failsafe_stdwriter(self, *args, **kwargs):
    try:
        self.stdwriter(*args, **kwargs)
    except BaseException as be:
        try:
            self._orig_file.write(*args, **kwargs)
            self._orig_file.write(f"\nFAILED WRITING WITH TIMESTAMP: {be!r}\n")
        except Exception:
            pass
        raise be


def setup_std_files():
#   sys.stdout, sys.stderr = open("stdout.log", "a+", buffering=1), open("stderr.log", "a+", buffering=1)
    for stream_name in ("stdout", "stderr"):
        sys_stream = getattr(sys, stream_name)
        f = open(f"{stream_name}.log", "a+")
        sys_stream_dup = os.dup(sys_stream.fileno())
        setattr(sys_stream, "_orig_file", open(sys_stream_dup, "w"))
        os.dup2(f.fileno(), sys_stream.fileno())

        setattr(sys_stream, "_last_write_ended_with_newline", True)
        setattr(sys_stream, "_orig_write", getattr(sys_stream, "write"))
        setattr(sys_stream, "write", MethodType(stdwriter, sys_stream))
        setattr(sys_stream, "stdwriter", MethodType(stdwriter, sys_stream))
        setattr(sys_stream, "write", MethodType(failsafe_stdwriter, sys_stream))

With the above changes, child outputs will also be written to the files. For example, if we replace 1/0 with subprocess.check_call(["ls", "/nosuch-dir"]) , the output will be as follows:

$ rm *.log; ./pyerr_subprocess.py ; for f in *.log; do echo "====== $f ====="; cat $f; echo "====== end ====="; done
====== stderr.log =====
ls: /nosuch-dir: No such file or directory
2023-01-14 08:22:41.945919: Traceback (most recent call last):
2023-01-14 08:22:41.945954:   File "/private/tmp/std_files/./pyerr_subprocess.py", line 80, in <module>
2023-01-14 08:22:41.946193:     run_some_failing_stuff()
2023-01-14 08:22:41.946232:   File "/private/tmp/std_files/./pyerr_subprocess.py", line 71, in run_some_failing_stuff
2023-01-14 08:22:41.946339:     raise e                             # to STDERR (only after `finally:`)
2023-01-14 08:22:41.946370:   File "/private/tmp/std_files/./pyerr_subprocess.py", line 68, in run_some_failing_stuff
2023-01-14 08:22:41.946463:     subprocess.check_call(["ls", "/nosuch-dir"])
2023-01-14 08:22:41.946494:   File "/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.9/lib/python3.9/subprocess.py", line 373, in check_call
2023-01-14 08:22:41.946740:     raise CalledProcessError(retcode, cmd)
2023-01-14 08:22:41.946774: subprocess.CalledProcessError: Command '['ls', '/nosuch-dir']' returned non-zero exit status 1.
====== end =====
====== stdout.log =====
2023-01-14 08:22:39.428945: hello
2023-01-14 08:22:39.428998: world
2023-01-14 08:22:39.429013: and.., whazzzuppp?
2023-01-14 08:22:41.931855: this line's timestamp should be ~2.5 seconds ahead of 'and.., whazzzuppp'
2023-01-14 08:22:41.945638: Exception handling: CalledProcessError(1, ['ls', '/nosuch-dir'])
2023-01-14 08:22:41.945774: Getting outta' here..
====== end =====

Children's output will not have a timestamp with their output (much like ls 's output in the above example), however this not only propagates to children, but also helps debugging (by allowing to fall-back to the original standard outputs). For example, a simple typo in stdwriter would be visible, as the output falls-back to the terminal:

$ rm *.log; ./pyerr_subprocess.py ; for f in *.log; do echo "====== $f ====="; cat $f; echo "====== end ====="; done
hello
FAILED WRITING WITH TIMESTAMP: NameError("name 'get_tamestamp_prefix' is not defined")

---

There might be better (and simpler) approaches. I did not explore the option of subclassing file types at all. The one thing I take away from this is - tinkering with error message output methods, makes for an "interesting" debug journey.